HomeTrace: the memory your household is missing

Amazon Developer Hackathon, Alexa+ track. A self-hosted MCP server (Streamable HTTP) that gives a household a voice-first, searchable memory of maintenance, warranties, repair history and promised follow-ups.

Nobody remembers when the RO filter was last changed, which appliance is still under warranty, or what the technician recommended. HomeTrace records it when it happens and retrieves it when someone asks.

Built during the hackathon window. The Alexa+ experience is simulated in a web app (no Alexa+ device needed).

Example interactions

• "The RO filter was changed today. The technician said to replace it again in six months."

• "When was the washing machine last repaired?"

• "Which warranties expire next month?"

• "Remind me to follow up with the electrician on Friday."

Related MCP server: HomeKeeper

MCP tools

The server computes all dates itself (for example "in six months"), so the LLM never does date maths. Data persists in data.json, so records survive restarts.

How it works

• src/mcp-server.js: the MCP server and its tools.

• server.js: serves the MCP endpoint at /mcp and the simulator backend. An MCP client lists the server's tools, Gemini decides which to call, and results flow back through MCP.

• public/index.html: the Alexa+-style simulator with voice input and spoken replies.

Quick start

Requirements: Node.js 20+ and a free Gemini API key from Google AI Studio.

git clone https://github.com/avneet-j/HomeTrace.git
cd HomeTrace
npm install

Create a .env file in the project root:

GEMINI_API_KEY=your_gemini_api_key
GEMINI_MODEL=gemini-2.5-flash
PORT=3000

Then run:

npm start

1. Open http://localhost:3000 in Chrome or Edge (voice input needs one of them).

2. Click Load demo history.

3. Type or speak the example interactions above.

Never commit your real .env file.

Use the MCP server directly

The endpoint is http://localhost:3000/mcp (Streamable HTTP, stateless).

To inspect it, run:

npx @modelcontextprotocol/inspector

Choose Streamable HTTP, enter the URL, click Connect, then List Tools.

Built with

Node.js, Express, Model Context Protocol SDK, Google Gemini API, vanilla HTML/CSS/JS.

Project status

HomeTrace is a functional hackathon prototype demonstrating conversational household management through Gemini-powered tool calling and an MCP server using Streamable HTTP.

The current implementation supports a single household, stores records locally in data.json, and does not include authentication or multi-user access controls. The Alexa+ experience is demonstrated through a web-based simulator rather than a native Alexa+ device integration.

Before public deployment, authentication, access controls, input validation, and secure data storage should be implemented and tested.

License

MIT, see LICENSE.